Christian | Tem-Tou.nft 🇭🇹

Anthropic engineers don't prompt Claude. They give it a job. So I tried it. A job description, a first week and a Friday performance review. Week one: graded B+, and it saved me roughly a full workday by my count. 1/ The job description: role, 6 duties, rules, who it reports to 2/ A brief every weekday morning, 8 lines max: what changed, what is stuck, what needs a decision 3/ Every request gets sorted: do now, draft for me, or ask me first 4/ Friday: it writes its own review, then I grade every duty from A to F The rule I'd steal first: "say where every number came from." That's how you check it. Agents are bad judges of their own work. Tara Research's OverclaimBench found agents misleading in 80.4% of incomplete runs. So when its grade and yours disagree, believe yours. How to copy it: start with one channel and one tool. Keep a duty only after two weeks of A or B. The full guide with every prompt is below ↓

‼️ Chinese Hackers Exploit VMware ESXi 0-Day to Hide Malicious Activity on Compromised Servers Details: cybersecuritynews.com/vmware… Chinese-linked hackers exploited a VMware ESXi zero-day vulnerability to hide activity on compromised servers during espionage operations against U.S. organizations, including law firms. CrowdStrike identified the group as WARP PANDA and said it used the flaw, tracked as CVE-2026-41709, in several intrusions, with exploitation continuing into early 2026. Broadcom’s security advisory describes CVE-2026-41709 as an insufficient logging vulnerability in VMware ESX. WARP PANDA was one of four prominent China-linked groups identified in the report. #cybersecuritynews